2 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
23 #include "xfs_trans.h"
26 #include "xfs_dmapi.h"
27 #include "xfs_mount.h"
28 #include "xfs_trans_priv.h"
29 #include "xfs_error.h"
31 STATIC
void xfs_ail_insert(struct xfs_ail
*, xfs_log_item_t
*);
32 STATIC xfs_log_item_t
* xfs_ail_delete(struct xfs_ail
*, xfs_log_item_t
*);
33 STATIC xfs_log_item_t
* xfs_ail_min(struct xfs_ail
*);
34 STATIC xfs_log_item_t
* xfs_ail_next(struct xfs_ail
*, xfs_log_item_t
*);
37 STATIC
void xfs_ail_check(struct xfs_ail
*, xfs_log_item_t
*);
39 #define xfs_ail_check(a,l)
44 * This is called by the log manager code to determine the LSN
45 * of the tail of the log. This is exactly the LSN of the first
46 * item in the AIL. If the AIL is empty, then this function
49 * We need the AIL lock in order to get a coherent read of the
50 * lsn of the last item in the AIL.
59 spin_lock(&mp
->m_ail_lock
);
60 lip
= xfs_ail_min(mp
->m_ail
);
66 spin_unlock(&mp
->m_ail_lock
);
74 * This routine is called to move the tail of the AIL forward. It does this by
75 * trying to flush items in the AIL whose lsns are below the given
78 * the push is run asynchronously in a separate thread, so we return the tail
79 * of the log right now instead of the tail after the push. This means we will
80 * either continue right away, or we will sleep waiting on the async thread to
83 * We do this unlocked - we only need to know whether there is anything in the
84 * AIL at the time we are called. We don't need to access the contents of
85 * any of the objects, so the lock is not needed.
90 xfs_lsn_t threshold_lsn
)
94 lip
= xfs_ail_min(mp
->m_ail
);
95 if (lip
&& !XFS_FORCED_SHUTDOWN(mp
)) {
96 if (XFS_LSN_CMP(threshold_lsn
, mp
->m_ail
->xa_target
) > 0)
97 xfsaild_wakeup(mp
->m_ail
, threshold_lsn
);
102 * AIL traversal cursor initialisation.
104 * The cursor keeps track of where our current traversal is up
105 * to by tracking the next ƣtem in the list for us. However, for
106 * this to be safe, removing an object from the AIL needs to invalidate
107 * any cursor that points to it. hence the traversal cursor needs to
108 * be linked to the struct xfs_ail so that deletion can search all the
109 * active cursors for invalidation.
111 * We don't link the push cursor because it is embedded in the struct
112 * xfs_ail and hence easily findable.
115 xfs_trans_ail_cursor_init(
116 struct xfs_ail
*ailp
,
117 struct xfs_ail_cursor
*cur
)
120 if (cur
== &ailp
->xa_cursors
)
123 cur
->next
= ailp
->xa_cursors
.next
;
124 ailp
->xa_cursors
.next
= cur
;
128 * Set the cursor to the next item, because when we look
129 * up the cursor the current item may have been freed.
132 xfs_trans_ail_cursor_set(
133 struct xfs_ail
*ailp
,
134 struct xfs_ail_cursor
*cur
,
135 struct xfs_log_item
*lip
)
138 cur
->item
= xfs_ail_next(ailp
, lip
);
142 * Get the next item in the traversal and advance the cursor.
143 * If the cursor was invalidated (inidicated by a lip of 1),
144 * restart the traversal.
146 STATIC
struct xfs_log_item
*
147 xfs_trans_ail_cursor_next(
148 struct xfs_ail
*ailp
,
149 struct xfs_ail_cursor
*cur
)
151 struct xfs_log_item
*lip
= cur
->item
;
153 if ((__psint_t
)lip
& 1)
154 lip
= xfs_ail_min(ailp
);
155 xfs_trans_ail_cursor_set(ailp
, cur
, lip
);
160 * Invalidate any cursor that is pointing to this item. This is
161 * called when an item is removed from the AIL. Any cursor pointing
162 * to this object is now invalid and the traversal needs to be
163 * terminated so it doesn't reference a freed object. We set the
164 * cursor item to a value of 1 so we can distinguish between an
165 * invalidation and the end of the list when getting the next item
169 xfs_trans_ail_cursor_clear(
170 struct xfs_ail
*ailp
,
171 struct xfs_log_item
*lip
)
173 struct xfs_ail_cursor
*cur
;
175 /* need to search all cursors */
176 for (cur
= &ailp
->xa_cursors
; cur
; cur
= cur
->next
) {
177 if (cur
->item
== lip
)
178 cur
->item
= (struct xfs_log_item
*)
179 ((__psint_t
)cur
->item
| 1);
184 * Now that the traversal is complete, we need to remove the cursor
185 * from the list of traversing cursors. Avoid removing the embedded
186 * push cursor, but use the fact it is alway present to make the
187 * list deletion simple.
190 xfs_trans_ail_cursor_done(
191 struct xfs_ail
*ailp
,
192 struct xfs_ail_cursor
*done
)
194 struct xfs_ail_cursor
*prev
= NULL
;
195 struct xfs_ail_cursor
*cur
;
198 if (done
== &ailp
->xa_cursors
)
200 prev
= &ailp
->xa_cursors
;
201 for (cur
= prev
->next
; cur
; prev
= cur
, cur
= prev
->next
) {
203 prev
->next
= cur
->next
;
211 * Return the item in the AIL with the current lsn.
212 * Return the current tree generation number for use
213 * in calls to xfs_trans_next_ail().
215 STATIC xfs_log_item_t
*
216 xfs_trans_first_push_ail(
217 struct xfs_ail
*ailp
,
218 struct xfs_ail_cursor
*cur
,
223 lip
= xfs_ail_min(ailp
);
224 xfs_trans_ail_cursor_set(ailp
, cur
, lip
);
228 list_for_each_entry(lip
, &ailp
->xa_ail
, li_ail
) {
229 if (XFS_LSN_CMP(lip
->li_lsn
, lsn
) >= 0) {
230 xfs_trans_ail_cursor_set(ailp
, cur
, lip
);
239 * Function that does the work of pushing on the AIL
243 struct xfs_ail
*ailp
,
246 long tout
= 1000; /* milliseconds */
247 xfs_lsn_t last_pushed_lsn
= *last_lsn
;
248 xfs_lsn_t target
= ailp
->xa_target
;
251 int flush_log
, count
, stuck
;
252 xfs_mount_t
*mp
= ailp
->xa_mount
;
253 struct xfs_ail_cursor
*cur
= &ailp
->xa_cursors
;
255 spin_lock(&mp
->m_ail_lock
);
256 xfs_trans_ail_cursor_init(ailp
, cur
);
257 lip
= xfs_trans_first_push_ail(ailp
, cur
, *last_lsn
);
258 if (!lip
|| XFS_FORCED_SHUTDOWN(mp
)) {
260 * AIL is empty or our push has reached the end.
262 xfs_trans_ail_cursor_done(ailp
, cur
);
263 spin_unlock(&mp
->m_ail_lock
);
268 XFS_STATS_INC(xs_push_ail
);
271 * While the item we are looking at is below the given threshold
272 * try to flush it out. We'd like not to stop until we've at least
273 * tried to push on everything in the AIL with an LSN less than
274 * the given threshold.
276 * However, we will stop after a certain number of pushes and wait
277 * for a reduced timeout to fire before pushing further. This
278 * prevents use from spinning when we can't do anything or there is
279 * lots of contention on the AIL lists.
283 flush_log
= stuck
= count
= 0;
284 while ((XFS_LSN_CMP(lip
->li_lsn
, target
) < 0)) {
287 * If we can lock the item without sleeping, unlock the AIL
288 * lock and flush the item. Then re-grab the AIL lock so we
289 * can look for the next item on the AIL. List changes are
290 * handled by the AIL lookup functions internally
292 * If we can't lock the item, either its holder will flush it
293 * or it is already being flushed or it is being relogged. In
294 * any of these case it is being taken care of and we can just
295 * skip to the next item in the list.
297 lock_result
= IOP_TRYLOCK(lip
);
298 spin_unlock(&mp
->m_ail_lock
);
299 switch (lock_result
) {
300 case XFS_ITEM_SUCCESS
:
301 XFS_STATS_INC(xs_push_ail_success
);
303 last_pushed_lsn
= lsn
;
306 case XFS_ITEM_PUSHBUF
:
307 XFS_STATS_INC(xs_push_ail_pushbuf
);
309 last_pushed_lsn
= lsn
;
312 case XFS_ITEM_PINNED
:
313 XFS_STATS_INC(xs_push_ail_pinned
);
318 case XFS_ITEM_LOCKED
:
319 XFS_STATS_INC(xs_push_ail_locked
);
320 last_pushed_lsn
= lsn
;
324 case XFS_ITEM_FLUSHING
:
325 XFS_STATS_INC(xs_push_ail_flushing
);
326 last_pushed_lsn
= lsn
;
335 spin_lock(&mp
->m_ail_lock
);
336 /* should we bother continuing? */
337 if (XFS_FORCED_SHUTDOWN(mp
))
344 * Are there too many items we can't do anything with?
345 * If we we are skipping too many items because we can't flush
346 * them or they are already being flushed, we back off and
347 * given them time to complete whatever operation is being
348 * done. i.e. remove pressure from the AIL while we can't make
349 * progress so traversals don't slow down further inserts and
350 * removals to/from the AIL.
352 * The value of 100 is an arbitrary magic number based on
358 lip
= xfs_trans_ail_cursor_next(ailp
, cur
);
363 xfs_trans_ail_cursor_done(ailp
, cur
);
364 spin_unlock(&mp
->m_ail_lock
);
368 * If something we need to push out was pinned, then
369 * push out the log so it will become unpinned and
370 * move forward in the AIL.
372 XFS_STATS_INC(xs_push_ail_flush
);
373 xfs_log_force(mp
, (xfs_lsn_t
)0, XFS_LOG_FORCE
);
377 /* We're past our target or empty, so idle */
379 } else if (XFS_LSN_CMP(lsn
, target
) >= 0) {
381 * We reached the target so wait a bit longer for I/O to
382 * complete and remove pushed items from the AIL before we
383 * start the next scan from the start of the AIL.
387 } else if ((stuck
* 100) / count
> 90) {
389 * Either there is a lot of contention on the AIL or we
390 * are stuck due to operations in progress. "Stuck" in this
391 * case is defined as >90% of the items we tried to push
394 * Backoff a bit more to allow some I/O to complete before
395 * continuing from where we were.
399 *last_lsn
= last_pushed_lsn
;
405 * This is to be called when an item is unlocked that may have
406 * been in the AIL. It will wake up the first member of the AIL
407 * wait list if this item's unlocking might allow it to progress.
408 * If the item is in the AIL, then we need to get the AIL lock
409 * while doing our checking so we don't race with someone going
410 * to sleep waiting for this event in xfs_trans_push_ail().
413 xfs_trans_unlocked_item(
417 xfs_log_item_t
*min_lip
;
420 * If we're forcibly shutting down, we may have
421 * unlocked log items arbitrarily. The last thing
422 * we want to do is to move the tail of the log
423 * over some potentially valid data.
425 if (!(lip
->li_flags
& XFS_LI_IN_AIL
) ||
426 XFS_FORCED_SHUTDOWN(mp
)) {
431 * This is the one case where we can call into xfs_ail_min()
432 * without holding the AIL lock because we only care about the
433 * case where we are at the tail of the AIL. If the object isn't
434 * at the tail, it doesn't matter what result we get back. This
435 * is slightly racy because since we were just unlocked, we could
436 * go to sleep between the call to xfs_ail_min and the call to
437 * xfs_log_move_tail, have someone else lock us, commit to us disk,
438 * move us out of the tail of the AIL, and then we wake up. However,
439 * the call to xfs_log_move_tail() doesn't do anything if there's
440 * not enough free space to wake people up so we're safe calling it.
442 min_lip
= xfs_ail_min(mp
->m_ail
);
445 xfs_log_move_tail(mp
, 1);
446 } /* xfs_trans_unlocked_item */
450 * Update the position of the item in the AIL with the new
451 * lsn. If it is not yet in the AIL, add it. Otherwise, move
452 * it to its new position by removing it and re-adding it.
454 * Wakeup anyone with an lsn less than the item's lsn. If the item
455 * we move in the AIL is the minimum one, update the tail lsn in the
458 * This function must be called with the AIL lock held. The lock
459 * is dropped before returning.
462 xfs_trans_update_ail(
465 xfs_lsn_t lsn
) __releases(mp
->m_ail_lock
)
467 xfs_log_item_t
*dlip
=NULL
;
468 xfs_log_item_t
*mlip
; /* ptr to minimum lip */
470 mlip
= xfs_ail_min(mp
->m_ail
);
472 if (lip
->li_flags
& XFS_LI_IN_AIL
) {
473 dlip
= xfs_ail_delete(mp
->m_ail
, lip
);
475 xfs_trans_ail_cursor_clear(mp
->m_ail
, dlip
);
477 lip
->li_flags
|= XFS_LI_IN_AIL
;
481 xfs_ail_insert(mp
->m_ail
, lip
);
484 mlip
= xfs_ail_min(mp
->m_ail
);
485 spin_unlock(&mp
->m_ail_lock
);
486 xfs_log_move_tail(mp
, mlip
->li_lsn
);
488 spin_unlock(&mp
->m_ail_lock
);
492 } /* xfs_trans_update_ail */
495 * Delete the given item from the AIL. It must already be in
498 * Wakeup anyone with an lsn less than item's lsn. If the item
499 * we delete in the AIL is the minimum one, update the tail lsn in the
502 * Clear the IN_AIL flag from the item, reset its lsn to 0, and
503 * bump the AIL's generation count to indicate that the tree
506 * This function must be called with the AIL lock held. The lock
507 * is dropped before returning.
510 xfs_trans_delete_ail(
512 xfs_log_item_t
*lip
) __releases(mp
->m_ail_lock
)
514 xfs_log_item_t
*dlip
;
515 xfs_log_item_t
*mlip
;
517 if (lip
->li_flags
& XFS_LI_IN_AIL
) {
518 mlip
= xfs_ail_min(mp
->m_ail
);
519 dlip
= xfs_ail_delete(mp
->m_ail
, lip
);
521 xfs_trans_ail_cursor_clear(mp
->m_ail
, dlip
);
524 lip
->li_flags
&= ~XFS_LI_IN_AIL
;
528 mlip
= xfs_ail_min(mp
->m_ail
);
529 spin_unlock(&mp
->m_ail_lock
);
530 xfs_log_move_tail(mp
, (mlip
? mlip
->li_lsn
: 0));
532 spin_unlock(&mp
->m_ail_lock
);
537 * If the file system is not being shutdown, we are in
538 * serious trouble if we get to this stage.
540 if (XFS_FORCED_SHUTDOWN(mp
))
541 spin_unlock(&mp
->m_ail_lock
);
543 xfs_cmn_err(XFS_PTAG_AILDELETE
, CE_ALERT
, mp
,
544 "%s: attempting to delete a log item that is not in the AIL",
546 spin_unlock(&mp
->m_ail_lock
);
547 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
555 * Return the item in the AIL with the smallest lsn.
556 * Return the current tree generation number for use
557 * in calls to xfs_trans_next_ail().
561 struct xfs_mount
*mp
,
562 struct xfs_ail_cursor
*cur
)
565 struct xfs_ail
*ailp
= mp
->m_ail
;
567 lip
= xfs_ail_min(ailp
);
568 xfs_trans_ail_cursor_set(ailp
, cur
, lip
);
574 * Grab the next item in the AIL from the cursor passed in.
578 struct xfs_mount
*mp
,
579 struct xfs_ail_cursor
*cur
)
581 struct xfs_ail
*ailp
= mp
->m_ail
;
583 return xfs_trans_ail_cursor_next(ailp
, cur
);
588 * The active item list (AIL) is a doubly linked list of log
589 * items sorted by ascending lsn. The base of the list is
590 * a forw/back pointer pair embedded in the xfs mount structure.
591 * The base is initialized with both pointers pointing to the
592 * base. This case always needs to be distinguished, because
593 * the base has no lsn to look at. We almost always insert
594 * at the end of the list, so on inserts we search from the
595 * end of the list to find where the new item belongs.
599 * Initialize the doubly linked list to point only to itself.
605 struct xfs_ail
*ailp
;
608 ailp
= kmem_zalloc(sizeof(struct xfs_ail
), KM_MAYFAIL
);
613 INIT_LIST_HEAD(&ailp
->xa_ail
);
614 error
= xfsaild_start(ailp
);
626 xfs_trans_ail_destroy(
629 struct xfs_ail
*ailp
= mp
->m_ail
;
636 * Insert the given log item into the AIL.
637 * We almost always insert at the end of the list, so on inserts
638 * we search from the end of the list to find where the
643 struct xfs_ail
*ailp
,
647 xfs_log_item_t
*next_lip
;
650 * If the list is empty, just insert the item.
652 if (list_empty(&ailp
->xa_ail
)) {
653 list_add(&lip
->li_ail
, &ailp
->xa_ail
);
657 list_for_each_entry_reverse(next_lip
, &ailp
->xa_ail
, li_ail
) {
658 if (XFS_LSN_CMP(next_lip
->li_lsn
, lip
->li_lsn
) <= 0)
662 ASSERT((&next_lip
->li_ail
== &ailp
->xa_ail
) ||
663 (XFS_LSN_CMP(next_lip
->li_lsn
, lip
->li_lsn
) <= 0));
665 list_add(&lip
->li_ail
, &next_lip
->li_ail
);
667 xfs_ail_check(ailp
, lip
);
672 * Delete the given item from the AIL. Return a pointer to the item.
675 STATIC xfs_log_item_t
*
677 struct xfs_ail
*ailp
,
681 xfs_ail_check(ailp
, lip
);
683 list_del(&lip
->li_ail
);
689 * Return a pointer to the first item in the AIL.
690 * If the AIL is empty, then return NULL.
692 STATIC xfs_log_item_t
*
694 struct xfs_ail
*ailp
)
697 if (list_empty(&ailp
->xa_ail
))
700 return list_first_entry(&ailp
->xa_ail
, xfs_log_item_t
, li_ail
);
704 * Return a pointer to the item which follows
705 * the given item in the AIL. If the given item
706 * is the last item in the list, then return NULL.
708 STATIC xfs_log_item_t
*
710 struct xfs_ail
*ailp
,
714 if (lip
->li_ail
.next
== &ailp
->xa_ail
)
717 return list_first_entry(&lip
->li_ail
, xfs_log_item_t
, li_ail
);
722 * Check that the list is sorted as it should be.
726 struct xfs_ail
*ailp
,
729 xfs_log_item_t
*prev_lip
;
731 if (list_empty(&ailp
->xa_ail
))
735 * Check the next and previous entries are valid.
737 ASSERT((lip
->li_flags
& XFS_LI_IN_AIL
) != 0);
738 prev_lip
= list_entry(lip
->li_ail
.prev
, xfs_log_item_t
, li_ail
);
739 if (&prev_lip
->li_ail
!= &ailp
->xa_ail
)
740 ASSERT(XFS_LSN_CMP(prev_lip
->li_lsn
, lip
->li_lsn
) <= 0);
742 prev_lip
= list_entry(lip
->li_ail
.next
, xfs_log_item_t
, li_ail
);
743 if (&prev_lip
->li_ail
!= &ailp
->xa_ail
)
744 ASSERT(XFS_LSN_CMP(prev_lip
->li_lsn
, lip
->li_lsn
) >= 0);
747 #ifdef XFS_TRANS_DEBUG
749 * Walk the list checking lsn ordering, and that every entry has the
750 * XFS_LI_IN_AIL flag set. This is really expensive, so only do it
751 * when specifically debugging the transaction subsystem.
753 prev_lip
= list_entry(&ailp
->xa_ail
, xfs_log_item_t
, li_ail
);
754 list_for_each_entry(lip
, &ailp
->xa_ail
, li_ail
) {
755 if (&prev_lip
->li_ail
!= &ailp
->xa_ail
)
756 ASSERT(XFS_LSN_CMP(prev_lip
->li_lsn
, lip
->li_lsn
) <= 0);
757 ASSERT((lip
->li_flags
& XFS_LI_IN_AIL
) != 0);
760 #endif /* XFS_TRANS_DEBUG */